Background: CRISPR-cpf1 is a single RNA-guided endonuclease system, becoming a promising tool in both prokaryotic and eukaryotic genome engineering. The editing efficiency of Cpf1 based engineering still requires improvements. However, limited information regarding the relationship between guide RNA sequence and on-target activity is available. To address these challenges, we developed a screening platform based on the association of Acidaminococcus sp. Cpf1(AsCpf1) DNA cleavage with cellular lethality. Major results: In total, we measured the activities of 12,544 guide RNAs, and observed a substantial variation of the editing efficiency depending on the design of the sequence. Based on this large-scale dataset, we designed and implemented a comprehensive computational model to predict activities of guide RNAs. Through comparison using simulated and experimental data, our approach outperformed existing algorithms, enabling selection of efficient guide RNAs. Conclusions: We refine on-target design rules and isolate the important sequence features that contribute to DNA cleavage, that is, AH dimers at position1-8 of protospacer promoting Cas12a activity while TK, GB dimer playing an inhibitory role. We validate guide RNA affinities designed by our optimized rules in both E.coli and 293T cells.